During a time where a master node is being upgraded, all data on live events running on a slave node is lost. Without protections keeping the slave node from connecting to the master database when the code versions are incompatible, the slave node could access an incompatible database that would cause the slave node to become unstable and crash.
It would be desirable to implement a handshaking methodology and a redundant data channel between servers during a software upgrade.